[distcc] Remote Fallback

[Martin Pool]

On 22 May 2003, Thomas Walker <Thomas.Walker at morganstanley.com> wrote:
> hmm... this is less trivial than I thought.  Since you handle localhost in such a
> different way (just exec the compiler rather than pass it all of the info from the cpp
> run) this gets complicated.  If, while looping through the host list looking for a
> good server, I hit localhost, it jumps out and runs "run_local" which in turn runs gcc
> (the whole thing, including cpp again) which isn't so good.  I assume the way
> "run_local" works was done just for simplicity?

There are two reasons:

If we originally choose to run locally then it is faster to run
cpp+cc1 in one go, rather than separately.

If we failed to distribute then it is possible that there is something
about the job that means splitting cpp from cc1 won't work, and so
it's safer to run the compiler in exactly the way the user asked.  I'm
not sure that will ever help very much, but since this is meant to be
a backup for a case where we already messed up it made sense to be as
conservative as possible.

If the problem was just that the remote machine wasn't available then
I don't think there's any need to run cpp again.

-- 
Martin